Anthropic study finds Claude exhibits different values depending on the language
Anthropic says an analysis of 300,000-plus anonymized Claude conversations found measurable shifts in how the model expresses warmth and rigor across languages.
Anthropic says a new study of more than 300,000 anonymized Claude conversations found that Claude expresses different values depending on the language being used, with higher warmth in Hindi and Arabic and more rigor in Russian. The company linked the claim to its research page, "Claude values across models and languages". On X, that quickly turned into an argument about what those shifts actually mean.
Combined views
156.3K
6 posts, first seen 9h ago